The field of predictive maintenance seeks to detect impending failures of components or processes. The detection is typically performed by first identifying of patterns in historical sensor data which preceded past failures, and then determining whether current sensor data conforms to the patterns. These techniques are unsuitable in cases where such historical sensor data, or training data, is not available in sufficient quantities.